OpenAI released GPT-6.1 Terra on 26 October, a mid-sized model that it says costs half as much as GPT-5.6 Terra did. It is available in ChatGPT Work, Codex and the API.
Developers who build on the API have the most to gain, provided the half-price claim holds for their own workloads. In ChatGPT Work the model is limited to paid plans.
Where Terra sits between Luna and Astra
OpenAI now sells four model classes, from cheapest to most capable: Luna, Terra, Sol and Astra. Terra was not part of last month’s GPT-6 release, which covered Sol and Luna, so it jumps a full generation, from 5.6 to 6.1. OpenAI’s launch post pitches it as quick and economical, and strong at understanding images and operating a computer, which is the kind of work behind agents that read a screen and click through apps.
Android Authority describes Terra as sitting between Sol and Luna: roughly half of Sol’s cost with similar performance, and sometimes close to Astra, which it calls OpenAI’s performance leader.
Error rates from OpenAI’s own test
On OpenAI’s internal benchmark, with reasoning effort set to low, the error rate went from 53.6% to 16.7%. With reasoning at maximum they dropped 18 points, to 4.3%. OpenAI also says Terra finishes computer-use tasks three times faster than its latest Astra and Sol models, and makes far fewer mistakes than the cheaper GPT-6 Luna. These are the company’s figures, not independent tests, and the publication’s benchmark chart carries no specific scores.
Who can use it
| Where | Status |
|---|---|
| ChatGPT Work | Plus, Pro, Enterprise and Edu users |
| Codex | Available |
| API | Available |
Per-token API prices were not in the sources we read, so the saving in dollars is still unknown, and “half” is a comparison with Terra’s own previous price rather than with Sol or Luna. For vision and computer-use jobs, such as reading screenshots or driving apps automatically, the cheaper model is worth testing against whatever you run today.
Independent benchmarks should start appearing within days, and they will show whether Terra holds up outside OpenAI’s own test. We’ll update this post if more details emerge.
